Nano Banana Pro vs Nano Banana 2.1 prompts: keep your Pro prompts, add a ratio and an identity line
We compared 65 Nano Banana Pro and 88 Nano Banana 2.1 prompts that creators posted on X, plus Google's own guide and prices. Google now recommends 2.1 for new projects at under half the price. In matched portrait edits, 81% of 2.1 prompts lock the face against 62% for Pro, and 56% name an aspect ratio against 31%.
Summary
If you have Nano Banana Pro prompts that work, move them to Nano Banana 2.1 as they are. Then add two things Pro prompts usually skip: an aspect ratio, and for edits, one sentence that locks the face.
The reason is simple: Google itself now points new work at 2.1. Its image generation guide calls 2.1 the efficiency workhorse and says it is recommended for all new projects. Per Google’s pricing, a 2.1 image costs $0.0336 at 1K, $0.0504 at 2K and $0.113 at 4K. A Pro image costs $0.134 at 1K or 2K and $0.24 at 4K.
Creators who moved to 2.1 write a little differently. We compared 65 Nano Banana Pro prompts and 88 Nano Banana 2.1 prompts that creators posted on X, each kept verbatim with its image:
- They name the shape. 53% of 2.1 prompts give an aspect ratio, against 23% for Pro. 2.1 adds the 1:4, 4:1, 1:8 and 8:1 ratios.
- They lock the face. In portrait edits from an uploaded photo, the most like-for-like task in both sets, 81% of 2.1 prompts tie the face or identity to keep or preserve, against 62% for Pro.
- They drop JSON. 32% of Pro prompts are written as JSON. For 2.1 it is 12%.
One caution up front: 63 of the 65 Pro prompts are Christmas and winter scenes posted in December 2025. That is why we lean on the portrait edit subset and on Google’s own pages. The details are under Limitations.
Our picks
- Default to 2.1, keep Pro for its three jobs. Google describes Pro as the premium choice for complex instructions, brand consistency and high-fidelity text. It also keeps more people consistent, 5 against 4, and takes up to 3 style references, which 2.1 does not. For everything else, 2.1 costs well under half as much per image.
- Name the aspect ratio. Over half of 2.1 prompts do, and the new tall and wide ratios only exist if you ask for them. The 2.1 prompts on our model page show each ratio with its output.
- Give the face its own sentence in an edit. 81% of 2.1 portrait edits have one. The fisheye café selfie starts a block with
IDENTITY LOCK: Preserve the exact facial identity of the reference person. Pro creators did this too, as in the theme park portrait that opens withDO NOT CHANGE THE FACE, just less often. - Number your uploads when you use several. Google’s template for combining images is
Take the [element from image 1] and place it with/on the [element from image 2]. None of the 65 Pro prompts and 2 of the 88 for 2.1 number their uploads. 2.1 takes up to 10 object images and 4 character images in one request, so saying which is which matters more as you add more. - Write sentences, or JSON if you prefer. The move away from JSON is real, but we have no evidence either format gives better images. All 27 prompt examples in Google’s guide are plain text.
Which of your Pro prompts are portrait edits? Those are the ones to check first: add the ratio, give the face its own sentence, and run them on 2.1.
How we tested
- The prompts. Every Nano Banana Pro and Nano Banana 2.1 entry in the open collection awesome-image-video-prompts at commit 590b532: 155 prompts. All 65 Pro prompts and 88 of the 2.1 prompts come from creators’ posts on X; the other 2 for 2.1 come from Reddit, which we left out.
- Posting days. X post IDs carry their creation time. Nano Banana Pro was released on November 20, 2025, and its posts run from November 22 to December 31. Nano Banana 2.1 was released on October 6, 2026, and its posts run from October 5 to November 2. Both sets come from the first six weeks after each release.
- Matched subsets. The sets differ in task: 63% of Pro prompts are image edits, against 35% for 2.1. So we also compare portrait edits only, 16 prompts on each side, and prompts with 100 to 1,000 likes, 34 for Pro and 36 for 2.1.
- Features. For each prompt we counted characters, words and lines, and tested patterns: JSON, labeled sections, lists, an aspect ratio, an output size like 4K, camera terms, negative instructions (no, avoid, without), fill-in blanks, what must stay unchanged, a face or identity lock, numbered uploads, and a holiday subject. The exact patterns are in the script.
- Google’s pages. Model descriptions, reference limits, new ratios and the multi-image template come from Google’s image generation guide; prices from the Gemini API pricing page. Both checked on November 7, 2026.
Results
Creators on X, all prompts
| Nano Banana Pro (65) | Nano Banana 2.1 (88) | |
|---|---|---|
| Image edits | 63% | 35% |
| Median length | 1,266 characters, 10 lines | 1,749 characters, 5 lines |
| Written as JSON | 32% | 12% |
| Labeled sections | 15% | 27% |
| Aspect ratio named | 23% | 53% |
| Output size named (4K, 2K) | 26% | 18% |
| Camera terms | 49% | 55% |
| Negative instructions | 35% | 56% |
| Says what must stay | 37% | 50% |
| Numbered uploads | 0% | 2% |
| Holiday subject | 97% | 5% |
Portrait edits from an uploaded photo
| Nano Banana Pro (16) | Nano Banana 2.1 (16) | |
|---|---|---|
| Median length | 1,688 characters | 2,134 characters |
| Face or identity lock | 62% | 81% |
| Says what must stay | 56% | 81% |
| Aspect ratio named | 31% | 56% |
| Negative instructions | 44% | 75% |
| Camera terms | 69% | 88% |
| Written as JSON | 31% | 6% |
Google’s specs
| Nano Banana Pro | Nano Banana 2.1 | |
|---|---|---|
| Google’s description | premium choice for complex instructions, brand consistency, high-fidelity text | efficiency workhorse, recommended for all new projects |
| Object reference images | up to 6 | up to 10 |
| Character reference images | up to 5 | up to 4 |
| Style reference images | up to 3 | none |
| Extra aspect ratios | 1:4, 4:1, 1:8, 8:1 | |
| Price per image | $0.134 (1K, 2K), $0.24 (4K) | $0.0336 (1K), $0.0504 (2K), $0.113 (4K) |
Are 2.1 prompts just longer because the set is different? The likes check says mostly yes. Among prompts with 100 to 1,000 likes, the medians are 1,437 characters for Pro and 1,495 for 2.1. The length gap is about which posts were collected; the ratio and face lock gaps hold in the portrait subset, where the task is the same.
What to ignore
- Length. Treat the overall gap of about 500 characters as a collection effect, not advice to write more. Prompts in the same likes band are the same length.
- Likes. The median is 105 for Pro and 78 for 2.1. The Pro posts were collected on October 29, 2026, ten months after they went up; most 2.1 posts were collected within three weeks. Comparing likes compares how long each post had to collect them.
- The 4K gap. Pro prompts ask for 4K a little more often, 26% against 18%. 2.1 renders 4K too, so this is a habit, not a capability.
Limitations
- The Pro set is seasonal. 63 of 65 Pro prompts are Christmas or winter scenes from December 2025. Subject drives some features, such as camera terms in portraits. The portrait edit subset narrows this gap but cannot remove it.
- Small subsets. The portrait comparison is 16 prompts per side, so one prompt moves a percentage by six points.
- We measured how people write. We did not run the same prompt on both models, so this says nothing about which image comes out better.
- The features are text patterns, written mostly for English. A ratio like
3:4counts; a prompt that says “make it tall” does not.
Reproduce it
The script is nano-banana-pro-vs-2-1-prompts-2026-11.py. Clone the public collection, check out commit 590b532 and run the script on it; it needs only Python and PyYAML. It writes nano-banana-pro-vs-2-1-prompts-2026-11.json: one row per prompt with its features, posting day and a link to the source file, plus every group summary in this post. Prompt text is not copied into the data; follow the link in each row.